[FEATURE] Add AI-Powered DevWeekends Assistant

[junaiddshaukat]

Feature Description

Add an AI-powered assistant to the DevWeekends platform that can answer questions based on DevWeekends data, including mission, events, sessions, community guidelines, mentor programs, blog posts, and general resources.
The assistant should follow a RAG (Retrieval Augmented Generation) approach using vector search and a lightweight LLM (e.g., Groq with Llama 3.1) to ensure accurate responses grounded in DevWeekends content.

Is your feature request related to a problem? Please describe.

New and existing community members often ask repetitive questions about sessions, joining rules, team info, upcoming events, and opportunities. This results in extra load on maintainers and ambassadors who need to manually respond.
There is no centralized place where users can quickly get accurate information based on DevWeekends’ own content.

Describe the solution you'd like
Introduce an AI assistant accessible from the DevWeekends website.
The assistant should:

Answer questions strictly based on DevWeekends’ data sources

Retrieve relevant content through vector search

Provide clear and consistent responses using an LLM

Allow easy updates when new events, blogs, or documents are added

Be lightweight enough to run on free-tier infra (Supabase + Groq)

The user experience should feel like a small chat widget embedded into the site.

Describe alternatives you've considered
Traditional FAQ page. Useful but limited, and users often ignore it.

Manual support via email or social media. Time-consuming and not scalable.

Search bar with keyword matching. Doesn’t understand context or long questions.

These alternatives don’t provide the conversational, context-aware experience that an AI assistant can.

Additional Information

Which part of the platform would this affect?

• Public website (home, sessions, mentors, etc.)
• Admin dashboard
• Mentor portal
• Ambassador portal
• API
• Other (please specify)

[Muhammad-AnasKhan]

Proposed high-level flow:

Content → Ingestion → Supabase (pgvector) → Query Embed → Retrieval → LLM → Answer

• Ingestion job indexes DevWeekends content into Supabase with embeddings.
• User asks a question via the chat widget.
• Backend embeds the question and retrieves the most relevant chunks from Supabase.
• Retrieved content + question are sent to a lightweight LLM (e.g. Groq / Llama 3.1, etc).
• Assistant returns an answer based on the retrieved DevWeekends content.

Feature Clarifications

DevWeekends data sources:

• What will we treat as the official sources for the assistant?
• For example: website pages, events/sessions data, community guidelines, blog posts, repo markdown, etc., and where do these live today (DB, CMS, repo, static pages)?

Embedding & indexing process:

• Proposed flow is to ingest approved content → split into chunks → generate embeddings → store them in Supabase Postgres (pgvector).
• Do we have a preferred embedding model for this (ideally free / open-source), or is the choice open as long as it’s lightweight?

Initial ingestion & updates:

• Should the initial embedding/indexing be handled via a manual script or an automated job?
• For v1, we could start with manual ingestion (run when data changes) and move to automation later as content updates become more frequent.

Query-time embeddings:

• Even after ingestion, the user’s question still needs to be embedded (using the same model) so we can find relevant chunks in Supabase.
• Should this embedding happen inside our backend (if supported) using a small open-source model, or via a hosted embedding API?